Method and device for adjusting neural-network-based wireless modem

Abstract

A method for adjusting a wireless modem includes: a channel parameter of a wireless modem at a present moment is acquired; a target clock frequency and a target working voltage of the wireless modem are generated, according to the channel parameter, with a neural network that is pre-trained; and a working voltage and a clock frequency of the wireless modem are adjusted to the target working voltage and the target clock frequency, respectively.

What is claimed is: 
     
       1. A method for adjusting a wireless modem, comprising:
 acquiring a channel parameter of the wireless modem at a present moment; 
 generating, according to the channel parameter, a target clock frequency and a target working voltage of the wireless modem with a neural network which is pre-trained; and 
 adjusting a working voltage and a clock frequency of the wireless modem to the target working voltage and the target clock frequency, respectively, 
 wherein the neural network is pre-trained through operations of: 
 acquiring a sample input set and an associated sample output set, each sample input of the sample input set comprising a group of channel parameters which are preset for the wireless modem, each sample output of the associated sample output set comprising a sample working voltage and a sample clock frequency of the wireless modem which operates under a corresponding group of channel parameters, and the sample working voltage and the sample clock frequency being a target working voltage and a target clock frequency of the wireless modem when a preset performance condition is met; and 
 training the neural network by using the group of channel parameters comprised in each sample input as the input of the neural network and by using the sample working voltage and the sample clock frequency comprised in the sample output associated with each sample input as output of the neural network. 
 
     
     
       2. The method of  claim 1 , wherein the acquiring the sample input set and the associated sample output set comprises:
 simulating, by a preset simulation software, a wireless channel corresponding to each group of channel parameters of a plurality of groups of channel parameters and simulating downlink data to be transmitted via the wireless channel; 
 simulating, by the simulation software, a situation in which the wireless modem receives the downlink data via the wireless channel, and determining the target working voltage and the target clock frequency of the wireless modem when the situation meets the preset performance condition; and 
 using each group of input channel parameters as a sample input, and using the target working voltage and the target clock frequency of the wireless modem which are outputted by the simulation software when the preset performance condition is met as a sample output corresponding to each sample input. 
 
     
     
       3. The method of  claim 1 , wherein, after the training the neural network, the method further comprises: correcting the neural network by:
 correcting the neural network according to the target clock frequencies and the target working voltages generated by the neural network for many times within a preset historical time period before the present moment and/or a plurality of load values of the wireless modem within the preset historical time period. 
 
     
     
       4. The method of  claim 1 , wherein the channel parameter comprises at least one of:
 a channel bandwidth, signal strength, a Signal Noise Ratio (SNR), Reference Signal Receiving Power (RSRP), an uplink/downlink resource scheduling strategy, a modulation mode, a coding mode, or a transport block size. 
 
     
     
       5. The method of  claim 1 , wherein the preset performance condition comprises one of: a bit error rate is less than a threshold value of the bit error rate, a packet loss rate is less than a threshold value of the packet loss rate, or a transmission success rate is greater than a threshold value of the transmission success rate. 
     
     
       6. The method of  claim 2 , wherein the training the neural network comprises operations of:
 using a first channel parameter as input of an initial neural network to acquire output of the initial neural network, the first channel parameter being any group of channel parameters of the plurality of groups of channel parameters; 
 comparing the output of the initial neural network a first sample output to correct a parameter of at least one neuron of the initial neural network, the first sample output being a sample working voltage and a sample clock frequency of the wireless modem which operates under the first channel parameter; 
 repeating the above two operations until the initial neural network meets a preset condition; and 
 using the initial neural network meeting the preset condition as the neural network. 
 
     
     
       7. The method of  claim 6 , wherein the preset condition comprises: when the input of the initial neural network is any of channel parameter of the sample input set, the output of the initial neural network is consistent with the sample output corresponding to the channel parameter of the sample output set, or a difference value between the output of the initial neural network and the sample output corresponding to the channel parameter of the sample output set is less than a preset threshold value. 
     
     
       8. A device for adjusting a wireless modem, comprising:
 a processor; and 
 a memory configured to store an instruction executable for the processor, 
 wherein the processor is configured to: 
 acquire a channel parameter of the wireless modem at a present moment; 
 generate, according to the channel parameter, a target clock frequency and a target working voltage of the wireless modem with a neural network which is pre-trained; and 
 adjust a working voltage and a clock frequency of the wireless modem to the target working voltage and the target clock frequency respectively, 
 wherein the processor is further configured to pre-train the neural network by: 
 acquiring a sample input set and an associated sample output set, each sample input of the sample input set comprising a group of channel parameters which are preset for the wireless modem, each sample output of the associated sample output set comprising a sample working voltage and a sample clock frequency of the wireless modem which operates under a corresponding group of channel parameters and the sample working voltage and the sample clock frequency being a target working voltage and a target clock frequency of the wireless modem when a preset performance condition is met; and 
 training the neural network by using the group of channel parameters comprised in each sample input as the input of the neural network and by using the sample working voltage and the sample clock frequency comprised in the sample output associated with each sample input as output of the neural network. 
 
     
     
       9. The device of  claim 8 , wherein the processor is configured to acquire the sample input set and the associated sample output set by:
 simulating, by preset simulation software, a wireless channel corresponding to each group of channel parameters of a plurality of groups of channel parameters and simulating downlink data to be transmitted via the wireless channel; 
 simulating, by the simulation software, a situation in which the wireless modem receives the downlink data via the wireless channel, and determining the target working voltage and the target clock frequency of the wireless modem when the situation meets the preset performance condition; and 
 using each group of input channel parameters as a sample input, and using the target working voltage and the target clock frequency of the wireless modem which are outputted by the simulation software when the preset performance condition is met as a sample output corresponding to each sample input. 
 
     
     
       10. The device of  claim 8 , wherein, after the training the neural network, the processor is further configured to correct the neural network by:
 correcting the neural network according to the target clock frequencies and the target working voltages generated by the neural network for many times within a preset historical time period before the present moment and/or a plurality of load values of the wireless modem within the preset historical time period. 
 
     
     
       11. The device of  claim 8 , wherein the channel parameter comprises at least one of:
 a channel bandwidth, signal strength, a Signal Noise Ratio (SNR), Reference Signal Receiving Power (RSRP), an uplink/downlink resource scheduling strategy, a modulation mode, a coding mode or a transport block size. 
 
     
     
       12. The device of  claim 8 , wherein the preset performance condition comprises one of: a bit error rate is less than a threshold value of the bit error rate, a packet loss rate is less than a threshold value of the packet loss rate, or a transmission success rate is greater than a threshold value of the transmission success rate. 
     
     
       13. The device of  claim 9 , wherein the training the neural network comprises operations of:
 using a first channel parameter as input of an initial neural network to acquire output of the initial neural network, the first channel parameter being any group of channel parameters of the plurality of groups of channel parameters; 
 comparing the output of the initial neural network a first sample output to correct a parameter of at least one neuron of the initial neural network, the first sample output being a sample working voltage and a sample clock frequency of the wireless modem which operates under the first channel parameter; 
 repeating the above two operations until the initial neural network meets a preset condition; and 
 using the initial neural network meeting the preset condition as the neural network. 
 
     
     
       14. The device of  claim 13 , wherein the preset condition comprises: when the input of the initial neural network is any of channel parameter of the sample input set, the output of the initial neural network is consistent with the sample output corresponding to the channel parameter of the sample output set, or a difference value between the output of the initial neural network and the sample output corresponding to the channel parameter of the sample output set is less than a preset threshold value. 
     
     
       15. A non-transitory computer-readable storage medium having stored therein a computer program instruction that, when being executed by a processor, implements the operations of a method for adjusting a wireless modem, wherein the method comprises:
 acquiring a channel parameter of the wireless modem at a present moment; 
 generating, according to the channel parameter, a target clock frequency and a target working voltage of the wireless modem with a neural network which is pre-trained; and 
 adjusting a working voltage and a clock frequency of the wireless modem to the target working voltage and the target clock frequency respectively, 
 wherein the neural network is pre-trained through operations of: 
 acquiring a sample input set and an associated sample output set, each sample input of the sample input set comprising a group of channel parameters which are preset for the wireless modem, each sample output of the associated sample output set comprising a sample working voltage and a sample clock frequency of the wireless modem which operates under a corresponding group of channel parameters, and the sample working voltage and the sample clock frequency being a target working voltage and a target clock frequency of the wireless modem when a preset performance condition is met; and 
 training the neural network by using the group of channel parameters comprised in each sample input as the input of the neural network and by using the sample working voltage and the sample clock frequency comprised in the sample output associated with each sample input as output of the neural network. 
 
     
     
       16. The non-transitory computer-readable storage medium of  claim 15 , wherein the acquiring the sample input set and the associated sample output set comprises:
 simulating, by a preset simulation software, a wireless channel corresponding to each group of channel parameters of a plurality of groups of channel parameters and simulating downlink data to be transmitted via the wireless channel; 
 simulating, by the simulation software, a situation in which the wireless modem receives the downlink data via the wireless channel, and determining the target working voltage and the target clock frequency of the wireless modem when the situation meets the preset performance condition; and 
 using each group of input channel parameters as a sample input, and using the target working voltage and the target clock frequency of the wireless modem which are outputted by the simulation software when the preset performance condition is met as a sample output corresponding to each sample input. 
 
     
     
       17. The non-transitory computer-readable storage medium of  claim 15 , wherein, after the training the neural network, the method further comprises: correcting the neural network by:
 correcting the neural network according to the target clock frequencies and the target working voltages generated by the neural network for many times within a preset historical time period before the present moment and/or a plurality of load values of the wireless modem within the preset historical time period. 
 
     
     
       18. The non-transitory computer-readable storage medium of  claim 15 , wherein the channel parameter comprises at least one of:
 a channel bandwidth, signal strength, a Signal Noise Ratio (SNR), Reference Signal Receiving Power (RSRP), an uplink/downlink resource scheduling strategy, a modulation mode, a coding mode, or a transport block size. 
 
     
     
       19. The non-transitory computer-readable storage medium of  claim 15 , wherein the preset performance condition comprises one of: a bit error rate is less than a threshold value of the bit error rate, a packet loss rate is less than a threshold value of the packet loss rate, or a transmission success rate is greater than a threshold value of the transmission success rate. 
     
     
       20. The non-transitory computer-readable storage medium of  claim 16 , wherein the training the neural network comprises operations of:
 using a first channel parameter as input of an initial neural network to acquire output of the initial neural network, the first channel parameter being any group of channel parameters of the plurality of groups of channel parameters; 
 comparing the output of the initial neural network a first sample output to correct a parameter of at least one neuron of the initial neural network, the first sample output being a sample working voltage and a sample clock frequency of the wireless modem which operates under the first channel parameter; 
 repeating the above two operations until the initial neural network meets a preset condition; and 
 using the initial neural network meeting the preset condition as the neural network.